Impact of computed tomography (CT) reconstruction kernels on radiotherapy dose calculation

Abstract Purpose To quantitatively evaluate the effect of computed tomography (CT) reconstruction kernels on various dose calculation algorithms with heterogeneity correction. Methods The gammex electron density (ED) Phantom was scanned with the Siemens PET/CT Biograph20 mCT and reconstructed with twelve different kernel options. Hounsfield unit (HU) vs electron density (ED) curves were generated to compare absolute differences. Scans were repeated under head and pelvis protocols and reconstructed per H40s (head) and B40s (pelvis) kernels. In addition, raw data from a full‐body patient scan were also reconstructed using the four B kernels. Per reconstruction, photon (3D and VMAT), electron (18 and 20 MeV) and proton (single field) treatment plans were generated using Varian Eclipse dose calculation algorithms. Photon and electron plans were also simulated to pass through cortical bone vs liver plugs of the phantom for kernel comparison. Treatment field monitor units (MU) and isodose volumes were compared across all scenarios. Results The twelve kernels resulted in minor differences in HU, except at the extreme ends of the density curve with a maximum absolute difference of 55.2 HU. The head and pelvis scans of the phantom resulted in absolute HU differences of up to 49.1 HU for cortical bone and 45.1 HU for lung 300, which is a relative difference of 4.1% and 6.2%, respectively. MU comparisons across photon and proton calculation algorithms for the patient and phantom scans were within 1–2 MU, with a maximum difference of 5.4 MU found for the 20 MeV electron plan. The 20MeV electron plan also displayed maximum differences in isodose volumes of 20.4 cc for V90%. Conclusion Clinically insignificant differences were found among the various kernel generated plans for photon and proton plans calculated on patient and phantom scan data. However, differences in isodose volumes found for higher energy electron plans amongst the kernels may have clinical implications for prescribing dose to an isodose level.


| INTRODUCTION
Computed tomography (CT) imaging is the current backbone of the entire radiotherapy treatment planning process. The scan(s) acquired during simulation set the stage for daily immobilization setup, target volume and organs-at-risk delineation, as well as treatment plan dose calculation. Aside from the importance of good image quality for contouring, the electron density information is immensely crucial for accurate radiation modeling of the dose delivered to the patient. A CT calibration curve converts the Hounsfeld unit (HU) values of different materials to electron density (ED) from which treatment planning dose calculation algorithms model the interactions of the incident radiation within the patient in order to calculate the dose.
This curve is typically defined during the initial stages of commissioning a treatment planning system. 1 This curve is specific to the CT scanner from which it is acquired and thus must be regenerated with the installation of any new CT scanners that will be used to image radiotherapy patients.
In the modern-day era of radiation therapy, the rapid progression of technology has introduced upgraded CT equipment with an abundance of features into the clinic. Some of these new features include metal artifact reduction, 2,3 extended field-of-view, 4 dual-energy imaging, [5][6][7] iterative reconstruction, 8 and automated tube voltage selection. 9 The role and impact of these new features has been and continues to be investigated and reported upon in the literature. The more traditionally customizable CT scan parameters such as kilovoltage, current, resolution, slice thickness, field-of-view (FOV) and reconstruction algorithm, have been more heavily studied in terms of the induced HU changes and subsequent impact on dose calculation. [10][11][12][13][14][15][16][17][18][19] These options can be varied through the selection of various anatomic scan protocols pre-installed onto the scanner. In addition, there is also a large variety of reconstruction kernel options available from the manufacturer to choose from. These kernels impact the resolution and apparent noise of the image, sharpening or smoothing the image depending on the kernel selected. However, currently there exists a lack of guidance on the recommended selection for clinical use, nor is there a thorough quantitative comparison of these different reconstruction kernels and their impact on dose calculation accuracy. It is therefore the purpose of this work to quantify the impact of CT simulation reconstruction kernels on a variety of radiation modalities and dose calculation algorithms with heterogeneity correction.

| MATERIALS AND METHODS
The Gammex electron density (ED) Phantom was initially scanned with the Siemens PET/CT Biograph20 mCT (Siemens Healthineers, Munich, Germany). The acquisition parameters were as follows: 120 kVp, 332 mAs, CareDose4D on, 2 mm slices, 0.8 pitch, and 500 mm field-of-view (FOV   Figure 3 displays HU vs ED calibration curves generated from the twelve different reconstruction kernels. It is evident that the overlapping data points suggest minimal differences were found amongst the kernel options for the extracted HU values for a majority of the T A B L E 1 Measured Hounsfeld Units (HU) of the Gammex phantom plug inserts scanned with the Siemens PET/CT Biograph20 mCT and reconstructed using four B kernels, four D kernels and four H kernels.   Looking at the volumes of specific isodose level distributions (V105%, V100%, V90%, V50%, V30% and V10%), the largest differences (ranging from 15.9 to 26.9 cc absolute differences) are seen in Table 5 for the hotter isodose levels (100% and 105%) amongst the different kernels of the eMC plans on the patient reconstructions. Relative differences of these isodose level volumes range from 5.5% to 44%. However, all of the lower isodose levels (V90%, V50%, V30% and V10%) are within a few ccs of each other across the kernel comparisons. The V105% for AAA also demonstrates an absolute difference 22.6 cc between B20s and B70s, but as a relative difference, is only 3.3%. The rest of the AAA results are within 1-2% of each other. Acuros and PCS plans all show a smaller degree of differences in volume between B20s and B80s scans relative to eMC, with a range of 0.5-8.9 cc. Performing the same experiment with the plans generated on the head vs pelvis scans of the phantom similarly resulted in large volume differences for the 20 MeV beam calculated with eMC (i.e.V90% was 4.28 cc for B40s vs 24.7 cc for H40s), but was not the case for the studied photon and proton plans ( Table 6). All of the lower isodose level volumes displayed minor differences amongst all of the treatment modalities.

| RESULTS
Based on the results discovered for the higher energy eMC plans listed in Tables 4 and 6, line profiles were plotted for comparison of the 20 MeV plans generated from the head and pelvis kernels in   Comparison of the calibration curves generated from the twelve different reconstruction kernels (B vs D vs H) showed minor differences in HU for the studied plug inserts of the Gammex phantom, with a difference of 40-55HU at the extreme ends of the density curve ( Fig. 3 and Table 1). Further evaluation also demonstrated minimal differences in calculated MU amongst the generated photon, electron and proton plans on these reconstructed patient scans ( Unlike for the AAA photon calculations, the higher energy electron beam differences in the V90% isodose volumes between the kernels may very well have implications for prescribing dose. This has the potential to induce errors for clinical practices that prescribe to a specific isodose volume, such as those used for boosting seromas in the breast after the completion of tangential photon treatment. The degree of error that may be induced for these types of boost plans warrants further investigation.
However aside from the higher energy electrons, the dose calculations were largely similar across the various reconstruction kernels for the different treatment modalities. This may have implications for selecting more optimal image reconstruction kernels per anatomic site allowing easier target or OAR delineation, improving image fusion, and improving the accuracy of automated contouring.
In fact, investigating the impact of different reconstruction kernels on the accuracy of automated segmentation algorithms is exactly the goal of future work.